DISCUSSION PAPER SERIES

IZA DP No. 12207

Lisette SwartWiljan van den BergeKaren van der Wiel

Do Parents Work More When Children Start School? Evidence from the Netherlands

MARCH 2019

Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but IZA takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity.The IZA Institute of Labor Economics is an independent economic research institute that conducts research in labor economics and offers evidence-based policy advice on labor market issues. Supported by the Deutsche Post Foundation, IZA runs the world’s largest network of economists, whose research aims to provide answers to the global labor market challenges of our time. Our key objective is to build bridges between academic research, policymakers and society.IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

Schaumburg-Lippe-Straße 5–953113 Bonn, Germany

Phone: +49-228-3894-0Email: publications@iza.org www.iza.org

IZA – Institute of Labor Economics

DISCUSSION PAPER SERIES

ISSN: 2365-9793

IZA DP No. 12207

Do Parents Work More When Children Start School? Evidence from the Netherlands

MARCH 2019

Lisette SwartCPB Netherlands Bureau for Economic Policy Analysis

Wiljan van den BergeCPB Netherlands Bureau for Economic Policy Analysis and Erasmus University Roterdam

Karen van der WielCPB Netherlands Bureau for Economic Policy Analysis and IZA

ABSTRACT

IZA DP No. 12207 MARCH 2019

Do Parents Work More When Children Start School? Evidence from the Netherlands*

When children start school, parents save time and/or money. In this paper, we empirically

examine the impact of these changes to the family’s budget constraint on parents’ working

hours. Labor supply is theoretically expected to increase for parents who used to spend

time taking care of their children, but to decrease for fulltime working parents because

of an income effect: child care expenses drop. We show that the effect of additional

time dominates the income effect in the Netherlands, where children start school

(kindergarten) for approximately 20 hours a week in the month that they turn 4. Using

detailed administrative data on all parents, we find that the average mother’s hours worked

increases by 3% when her youngest child starts going to school. For their partners, who

experience a much smaller shock in terms of time, the increase in hours worked is also

much smaller at 0.4%.

JEL Classification: J13, J22

Keywords: labor supply, starting school, child care

Corresponding author:Wiljan van den BergeCPB Netherlands Bureau for Economic Policy AnalysisP.O. Box 805102508 GM The HagueThe Netherlands

E-mail: w.van.den.berge@cpb.nl

* We thank Arthur van Soest, Jan van Ours, Egbert Jongen, Pierre Koning, Dinand Webbink and seminar participants

at EALE in Lyon and CPB for comments and suggestions. Remaining errors are our own.

1 Introduction

The first day a child enters school is a very exciting day in his or her life: it marks the

start of many new experiences and the acquisition of many new skills. As economists

put it: children formally start investing in their human capital, and this process will

continue to influence their lives for many years to come. The benefits of going to school

on child development have been studied extensively: compulsory schooling does not

only improve test scores, but also causally enhances later life outcomes, such as wages

and health (Devereux and Hart, 2010; Machin et al., 2011; Grenet, 2013).

At the same time, compulsory schooling also provides benefits for parents, as they

start saving time and/or money – at least in countries where the government subsidizes

primary schools more than child care – which may affect their labor supply. On the

one hand, parents who used to take care of their children during school hours are

expected to increase their labor supply when their youngest child starts school. These

parents do not need to take care of their children anymore and hence have free time

on their hands. On the other hand, parents whose children attended (paid) childcare

before going to school might decrease their labor supply, when their youngest child

starts school. These households save on childcare expenses and therefore experience

an income effect.

In this paper, we empirically estimate the effect of children starting compulsory

schooling in the Netherlands on their parents’ labor market position. Using a balanced

panel of administrative data on all Dutch parents between 2006 and 2016, we analyze

the labor market position of parents when their youngest child is between three and

six years old, and we do this for mothers and fathers separately. We apply a difference-

in-differences approach to tease out any changes observed due to macro-level shocks,

increased working experience over these three years or other changes unrelated to

school. The control group consists of parents with a youngest child between one and

four years old. They therefore do not experience the ‘treatment’ because their youngest

child does not start school during the observation period. To ensure that our control

and treatment group are as similar as possible and to take cohort effects into account,

we apply coarsened exact matching.

How compulsory schooling affects the parents’ labor market position is of interna-

tional interest as employment of mothers of young children is relatively low in most

countries, and policy makers are hence interested in the drivers of their labor supply.

Interestingly, Figure 1 shows that in most European countries the differences in the

2

average number of hours worked by mothers with or without all their children in school

are small. It is unclear however to what extent this is driven by cohort effects: younger

cohorts with smaller children tend to work more in general.

The Netherlands is a particularly interesting country to study labor market effects

for parents of compulsory schooling. Primary education is paid for by the government,

while child care is not, and child care subsidies are dependent on household income.

Moreover, children start school (kindergarten) for approximately 20 hours a week in the

month that they turn four, which makes it possible to rule out seasonal employment

effects. Furthermore, the participation of Dutch women is relatively high, while their

working hours are relatively low compared to that of women in other European coun-

tries as the majority of Dutch women work part-time (see Figure 1). This means that

there is substantial room for increases in hours worked. A large majority of the men

however, works fulltime: in 2016, the employment rate among men was 82.6 percent

and more than three quarters works fulltime.

Our findings show economically small labor supply reactions and that mothers

adapt their working hours significantly more than fathers when their youngest child

starts school. On average, households save approximately 60 to 70 euros per month

on child care. Dutch mothers experience an increase in their free time of 13 hours

a week when their youngest child goes to school. After two years, we find that the

average number of hours Dutch mothers work increases by around 0.5 hours. These

changes are driven for about two-thirds by responses at the intensive and one-thirds at

the extensive margin. Given that the average mother in our sample works around 15

hours a week, mothers increase their work hours by 3% after their youngest child goes

to school. Empirically, the income effect is thus dominated by the effect of additional

time on labor supply. Dutch fathers experience an increase in their free time of almost

4 hours a week when their youngest child goes to school. We find that they increase

working hours on average by around 0.15 hours, or 0.4% relative to the mean. This is

for about two-thirds driven by a response along the extensive margin and one-thirds

at the intensive margin.

Heterogeneity analyses suggest that the response is larger among mothers and fa-

thers who are already working longer hours and who already earn higher wages. This

is perhaps surprising because they experience smaller time gains and they typically

save more money on child care expenses when their youngest child starts school. The-

oretically, one would thus expect smaller increases in working hours in these groups.

3

4

Figure 1: Hours worked by mothers before and after the youngest child in a householdstarts going to school

Source: Own calculations based on Eurostat (2014).

Notes: The x-axis indicates how many hours women work when the youngest child in their household

is older than three years old, but does not yet go to school and the y-axis denotes the number of hours

women work when the youngest child in their household goes to school and is younger than twelve

years old. These averages take all mothers in a country into account: mothers who do not work, work

zero hours. In addition, this figure is based on cross-sectional data, so that differences may also be

affected by cohort effects: overall labor participation of younger generations of women exceeds that of

older generations. Finally, the solid line denotes the diagonal: countries in which mothers of children

who do not go to school work as much as mothers whose children do go to school, are on the diagonal.

In countries above (below) the diagonal, women whose youngest child starts going to school work

longer (shorter) hours.

However, differences in initial preferences for labor supply over leisure are probably at

the heart of these results: these parents are more likely to have decreased working hours

solely to be with their children and hence return to the office once child care activi-

ties are less needed. Single mothers show an overall negative employment response to

their youngest child attending school. The income effect apparently dominates for this

subgroup. One explanation for the relatively large income effect is mental accounting

(Thaler, 1985; Abeler and Marklein, 2016). In the Netherlands, child care subsidies

depend on household income, are paid by the tax authorities, and parents generally

receive them on a different date than when they pay the child care institution. It is

therefore possible that parents do not take the net monetary gain they obtain when

their youngest child starts going to school into account, but rather consider the gross

monetary gain of reduced expenses on child care without taking the corresponding

decline in child care subsidies into account. This could lead to much larger perceived

income gains. The difference between the gross and the net monetary gain is generally

substantial; the gross monetary gain can be up ten times as large as the net monetary

gain for low-income households.

Our analysis makes several contributions to the literature on compulsory schooling

and the labor supply of parents, which generally uses data on earlier cohorts of parents.

The first contribution of this paper is that findings are not confounded by seasonal

effects in labor supply. In the Dutch institutional setting, children go to school on the

day they turn four years old. This provides us with variation in school entry over the

calendar year, unlike other studies from countries where children always start school

in September (Gelbach, 2002; Goux and Maurin, 2010; Finseraas et al., 2016). The

second contribution is that we can precisely estimate the effects using very similar

treatment and control groups. This is because we use a recent administrative dataset,

which includes the hours a recent cohort of parents work in each month as well as

information on child care subsidies for the universe of Dutch parents. In addition,

the data also enable us to determine the heterogeneity in the magnitude of the shock

parents experience in terms of money and time depending on their working hours

and the number of hours their youngest child attends formal child care. The third

contribution is that we symmetrically estimate effects for both mothers and fathers,

while most other papers in the literature focus on the employment effects of children

going to school for mothers only.

Interestingly, our heterogeneity analyses provide different results than most of the

5

literature; often older papers find stronger effects for single women and low-income

groups, but we draw opposite conclusions. A paper by Gelbach (2002) for example

examines the effect of public schooling for five year olds in the US on their mothers’

labor supply and finds that single mothers increase the number of hours they work by

6-24 percent, while for married mothers he finds an increase of 6-15 percent. Gelbach

uses cross-sectional data from the 1980 US Census, while we use panel data from 2006

to 2016. Goux and Maurin (2010) analyze census data from 1999 in France where all

children start school at age 3 or earlier using a cut-off within the year for eligibility

and find that pre-elementary school only has a significant effect on the labor supply

of single mothers. Similarly, Finseraas et al. (2016) examine a reform in Norway in

1997 where the compulsory school starting age was lowered from six to five. They

find a short-term increase of labor supply, with much stronger effects for mothers

with low wage potential, who probably did not use childcare. Recent cohorts differ

in important dimensions from older cohorts: they are higher educated, employment

rates have increased steadily (Blundell et al., 2013) and labor supply elasticities are

currently smaller (Blau and Kahn, 2007).

Another related branch of the literature looks at the effect of the availability of –

non-compulsory – kindergarten on the labor market position of parents. Cascio (2009)

exploits the introduction of subsidies for kindergarten in school districts in the US

from the 1960s to the 1980s to examine effects of public schooling on labor supply

of mothers. Using cross-section data from the Census for each decade from 1950 to

1990 she only finds significant effects for single mothers. Fitzpatrick (2010) considers

pre-kindergarten availability in three states in the US. Using Census data from 2000 in

an RD design, she finds no robust impact of pre-kindergarten availability on maternal

labor supply. Fitzpatrick (2012) uses the same data, but exploits cutoffs for eligibility

for public kindergarten. She finds that eligibility only increases the employment of

single mothers without additional children. Related, Shure (2019) considers the effect

of the extension of the schoolday in Germany in the 2000s on maternal labor supply. She

finds no effect along the intensive margin, but an increase in employment probability.

2 Theoretical Framework

This section formalizes the intuition of how parental labor supply is affected when

children start going to school into testable hypotheses. To do this, we analyze the

6

changes in the budget constraint parents face that occur when their youngest child

starts going to school. For simplification, we restrict our focus to three activities

parents can spend their working hours on1: (1) working on the labor market li, (2)

taking care of children at home ti or (3) leisure zi. The time constraint for each

individual i is therefore:

li + ti + zi = 1 where 0 ≥ li, ti, zi ≥ 1 (1)

Children always have to be taken care of. This can either be done by the father tf

or the mother tm, or by child care providers ty. Older children also spend a number of

hours at school, ts. As long as at least one child in a household does not go to school

yet, parents have to arrange child care during school hours. The time constraint of

taking care of the children can therefore be characterized by:

tf + tm + ty + ts = 1 where ts = 0 if at least one child does not go to school (2)

0 < ts < 1 if all children go to school.

Households earn income from labor li and spend money on child care ty which costs

q and on consumption X which costs p. Households therefore face the following budget

constraint:

qty + pX ≤ wf lf + wmlm (3)

In this budget constraint people cannot save any money for future consumption. Sub-

stituting the time constraints for fathers and mothers (Equation 1) as well as the time

constraint for child care (Equation 2) into this equation yields the following budget

constraint:

q(1− tm − tf − ts) + pX ≤ wf (1− tf − zf ) + wm(1− tm − zm) (4)

When the youngest child in a household starts going to school, the budget constraint

parents face changes. Figure 2 graphically illustrates the changes one parent faces,

keeping the time the other parent spends on labor lf and child care tf constant. In the

Figure, we show the mother’s budget constraint, but the budget constraints fathers

1The time available during the (work) week is normalized to 1 without loss of generality. Outsideof working hours parents can still spend time with their children and enjoy leisure.

7

face keeping the time mothers spend on labor and child care constant is identical.

Mothers decide how to allocate their time between working which enables them to

consume X, leisure zm and taking care of their children tm. This three-dimensional

budget constraint mothers face is shown in three two-dimensional views in Figure 2.

The blue solid planes represent the situation when at least one child does not go to

school yet and the green dashed planes represent the situation after the youngest child

has started going to school and parents no longer have to arrange child care during

school hours.

When the youngest child goes to school, parents do not need to pay for child care

for the hours during which the children are at school. Parents who pay for child care

therefore experience an income effect. This income effect shifts the budget constraint

upward and causes mothers to decrease their labor supply. At the same time, the

maximum amount of time parents can spend on domestic child care reduces, because

children do not need to be taken care of by their parents when they are at school.

This ‘time effect’ causes parents who used to take care of their children during school

hours prior to the youngest child going to school, to increase their labor supply. This

therefore yields the following hypotheses, which are not mutually exclusive:

Hypothesis 1 When the youngest child in the household starts going to school, work-

ing mothers experience an income effect, which decreases their labor supply.

Hypothesis 2 When the youngest child in the household starts going to school, moth-

ers who do not work fulltime obtain additional time, which increases their labor supply.

This theoretical framework also allows for the prediction of some heterogeneous

effects. If the real wage of the mother increases ceteris paribus, she experiences both a

substitution and an income effect. The income effect decreases labor supply, while the

substitution effect has an upward effect. Generally, the substitution effect dominates

the income effect. For higher real wages of the mother, the negative effect on maternal

labor supply is therefore smaller. Additionally, if the price of child care decreases

ceteris paribus, the marginal benefit of working increases, making it less attractive to

decrease labor supply.

In sum, when the youngest child starts going to school, this may affect maternal

labor supply in two ways. On the one hand, parents whose children attend external

child care experience an income effect that may reduce labor supply. On the other

8

Figure 2: The mother’s budget constraint.

(a) Domestic child care tm vs leisure zm (b) Consumption X vs leisure zm

(c) Consumption X vs child care tm

Legend

Situation lm zm tm X

1 1 0 0 w∗f lf + w∗m − q∗(1− ts − tf )

2 0 1 0 w∗f lf − q∗(1− ts − tf )

3A ts + tf 0 1− ts − tf w∗f lf + w∗m(ts + tf )

3B 0 ts + tf 1− ts − tf w∗f lfNotes: This Figure shows three two-dimensional views of the household budget constraint that parents face when

deciding on how to allocate their time. The blue solid planes represent the situation before the youngest child goes

to school, where ts = 0, while the green dashed planes denote the situation after the youngest child started school,

ts > 0. In the figures, situations before the child starts going to school are denoted by NSi and situations once all

children in a household go to school are referred to as Si. The numbers in the subscript refer to situations shown in the

legend. Appendix Section A.1 explains in more detail how this budget constraint changes when the youngest child in a

household starts going to school. Although the highest point of the budget constraint before children go to school (the

blue solid plane) in Figure 2c is higher than the kink of the budget constraint after children go to school (the green

dashed plane), these relative positions could be reversed without affecting the hypotheses.

hand, parents who provide domestic child care spend less time doing so which may

increase their labor supply.

3 Magnitude of the treatment

This section discusses the changes parents experience when the youngest child in a

household starts going to school. The magnitude of these changes determines the

treatment parents experience. This magnitude depends on the number of hours par-

ents work, the number of hours children attend formal child care and on the price of

child care. We start this section by explaining the institutional setting in the Nether-

lands concerning school and child care. Subsequently, we calculate the (counterfactual)

magnitude of the change mothers and fathers experience in terms of time when their

children start going to school, followed by a calculation of the magnitude of the change

households experience in terms of money. To calculate these shocks, we determine how

large the change would have been if the parental labor supply and the number of hours

children attend external child care had remained the same.

3.1 Institutional setting

In the Netherlands, children generally start school the day they turn four years old.

Although schooling is only compulsory from the age of five, more than 99 percent of

children starts school when they are four years old (Eurostat, 2012). In the first years

of primary school, children attend school for on average 22 hours a week, or on average

for 4.4 hours a day.2

Formal child care in the Netherlands is privatized and institutions can set their

own price.3 Parents can obtain a child care subsidy from the government for the hours

children between the age of 0 and 12 attend formal child care.4 This subsidy is a

percentage of the hourly price child care institutions charge. The subsidy varies with

household income: the government pays a lower share of the costs of child care for

2The number of hours children go to school varies between schools, because the only requirement setby the government is that in the first four years of primary school, children need to receive educationfor at least 3,520 hours (Rijksoverheid, 2017). This is 22 hours a week on average.

3The government annually sets a threshold price per hour and only for the amount parents paybelow this threshold they receive government subsidies.

4Formal child care includes day care for children younger than four years old, out-of-school carefor children in primary school and childminders for children between 0 and 12 years old.

10

households with a higher income than for households with a lower income. Primary

school on the other hand is paid for by the government, although schools may ask for

a ‘voluntary parental contribution’ which may range from 10 euros a year to 120 euros

a year.

3.2 Magnitude of the change in terms of time

This subsection calculates the number of hours parents save when their youngest child

starts going to school. Parents only save time when they do not work fulltime and

provide domestic child care during working hours. When the youngest child in a family

starts going to school, parents no longer need to take care of the children during school

hours.

Fulltime employment in the Netherlands takes up forty hours a week. A regular

working day therefore consists of eight hours and children go to school for on average

4.4 hours a day in the first years of primary school. The shock parents experience in

terms of time therefore consists of 4.4 hours a day for each day that a parent does not

work, in other words:

Time shock = (40− hours worked)× 4.4

8(5)

In these calculations, we assume that working hours are fixed to eight hours a day.

This implies that our calculation of the time shock is in fact an upper bound. Parents

who have flexible working hours may experience a shock smaller than the one calculated

here.

On average mothers save a substantial 13.3 hours per week when their youngest

child starts going to school, as illustrated in Column 2 of Panel A in Table 1. Column 2

of Panel B illustrates that the shock fathers experience in terms of time is substantially

smaller, only 3.7 hours on average. Fathers generally save less time because a much

larger share of them works fulltime, as is illustrated in Column 1 of Table 1. Parents

who work fulltime hardly save any time when their youngest child goes to school, as

they generally do not take care of the children during working hours before the children

start going to school. The time shock is largest for parents who work fewer hours.

11

Table 1: Magnitude of the shock in terms of money and in terms of time

Panel A: Shock in terms of time and money that mothers experience

when their youngest child goes to school

(1) (2) (3)

% of Hours saved Money saved

mothers per week per month

(per household)

Does not work 25.6 22.0 8

Works less than 20 hours a week 30.0 14.9 39

Works 20-34 hours a week 39.6 7.9 103

Works more than 34 hours a week 4.7 1.5 135

Total 100.0 13.3 61

Panel B: Shock in terms of time and money that fathers experience

when their youngest child goes to school

% of Hours saved Money saved

fathers per week per month

(per household)

Does not work 6.4 22.0 22

Works less than 20 hours a week 4.3 16.1 43

Works 20-34 hours a week 16.6 5.2 83

Works more than 34 hours a week 72.7 1.0 72

Total 100.0 3.7 70

Notes: See main text for calculation of the monetary and time shocks. The percentages listed in

Column (1) refer to the share of mothers (fathers) in our sample that works a specific number of

hours. The amounts of money households save when their youngest child starts going to school

(listed in Column (3)) differs for the mothers (Panel A) and fathers (Panel B) in our sample,

because households with single mothers are included in the sample for mothers, but not in the

sample of fathers.

3.3 Magnitude of the change in terms of money

Parents may also save money when their children start going to school. This only holds

true for parents whose children attend child care, as they do not need to pay for child

care during the hours the child is at school. When children start school, the number of

hours they attend formal child care therefore reduces by approximately fifty percent.5

5Child care providers generally charge parents for ten hours per day for children who have not yetstarted going to school yet and for five hours per day for out-of-school care for children who attend

12

The net income-shock parents experience due to reduced expenditures on formal child

care can therefore be approximated using the following formula:

Monetary shock = pNqN − pOqO = (sPN)(50%qO)− (sPO)qO = sqO (0.5PN − PO)

(6)

where subscript O denotes the old situation (day care) and N denotes the new situation

(out-of-school care). Additionally, p denotes the price parents pay for an hour of formal

child care, which is a share s of the actual hourly price of child care P : p = sP .6 The

price of out-of-school care pN may differ from the price of day care pO.7 Finally, q

refers to the number of hours a child attends formal child care, where the number of

hours in out-of school care qN is half the number of hours in the day care, qN = 0.5qO.8

On average, households save approximately 60 to 70 euros a month on the costs of

formal child care when their youngest child starts going to school, as shown in Column

3 of Table 1. Panel A shows that households where mothers work more hours save more

money in absolute terms than households where mothers work fewer hours per week,

because household income is generally higher - and childcare subsidies are therefore

lower - in households where mothers work more hours.9 Panel B shows that for fathers

primary school.6For households where both parents work or are enrolled in education, the government subsidizes

formal child care, which includes day care, out-of-school care and childminders for children between0 and 12 years old. The subsidy is a percentage of the price that depends on household income. Theshare paid by the government varied substantially over the past decade. Formal child care in theNetherlands is privatized and institutions can set their own price. The government annually sets athreshold price per hour set and only for the amount parents pay below this threshold they receivegovernment subsidies.

7For these hourly prices of formal child care, we use the average price reported by the sectorassociation of child care providers in the Netherlands in a specific year (Kinderopvang, 2016). Theshare paid by parents is calculated using a conversion table from the Dutch government that specifiesthis share for each calendar year for each level of household income.

8The hours a child attends formal child care are available per calendar year. We therefore use thenumber of hours in the calendar year in which the child turns three years old, i.e. the calendar yearbefore the child starts going to school.

9To be eligible for child care subsidies in the Netherlands, both parents need to work or be enrolledin education. Nonetheless, Table 1 shows that women who do not work (those working zero hours aweek) on average still save some money when their children start going to school. This is likely dueto mothers who are enrolled in education or mothers who recently became unemployed and who aretherefore still eligible for child care subsidies. Furthermore, before the reform of 2011, the eligibilityfor subsidies for formal child care did not depend on the number of hours parents worked. Therefore,before 2011 children of women who did not work could also attend formal child care so that thesewomen also experienced a small monetary shock the moment their youngest child started primaryschool.

13

this relationship is not linear: households where the fathers work 20-34 hours save

more money than households where fathers work more, because on average women

work more hours in households where fathers work 20-34 hours.

It is important to note that these average changes in the money parents spend

on child care when their children go to school are a lower bound the true magnitude

of the monetary shock. Our calculation only takes the costs of formal child care into

account, as there are no data for all parents on the use of informal child care. If parents

use paid informal child care, the true magnitude of the monetary shock exceeds our

calculations.

4 Empirical Approach

To determine the effect of children going to school on parental labor supply, we employ

a difference-in-differences design. We construct a control group of parents with children

aged one to four to control for general business cycle effects, policy changes and other

time-changing variables. To ensure comparability of the treatment and control group,

we employ coarsened exact matching and we do the analyses separately for mothers

and for fathers.

4.1 Defining treatment and control group

Our treatment group consists of parents whose youngest child is between three and six

years old. Since children go to school when they are four years old, this means that

a parent is in our sample one year before the youngest child goes to school until two

years after the child goes to school. To ensure that parents do in fact experience a time

shock when their youngest child goes to school, we impose that the child remains the

youngest child in the family until they are six years old. We end up with a balanced

sample that consists only of parents who took care of this particular child for each of

the three years in the sample.

To estimate the effect of the youngest child going to school on parental labor sup-

ply, we compare this treatment group to a control group consisting of parents whose

youngest child is between one and four years old. In this group the youngest child

does not start going to school yet during the observation period, so that parents do

not experience the treatment.10 With this design, we compare parents whose youngest

10Older children of parents in the control group may still start going to school in the observation

14

child turns four years old in a given year (the treatment group) with parents whose

youngest child turns two years old in that same year (the control group).

In the robustness checks we also present results where the control group consists

of parents whose second-youngest child is between three and six years old. This group

does not experience a time shock, since they still have a younger child at home, but

they do experience a small monetary shock, since they pay for fewer hours of child care

for their second-youngest child when this child starts going to school.11

4.2 A difference-in-differences design

We estimate the effect of children going to school on the number of hours worked by the

parent in a certain month t (those who do not work, work zero hours). We then check

whether this effect is driven by changes along the intensive or the extensive margin by

estimating the effect of children going to school on i) a dummy variable that indicates

whether the parent works in month t, and ii) the number of hours worked by the parent

in month t, excluding those who do not work over the entire observation period. We

estimate the following linear model

yit = α + β × Ti +24∑

t=−10

γt × St +24∑

t=−10

δt × St × Ti + ηi + θt + εit, (7)

where yit is some measure of labor market participation (hours worked or participation).

Ti is a dummy variable that takes value 1 if parent i is in the treatment group and 0

if the parent is in the control group. St is a set of dummy variables for each month

relative to the treatment period, where St = 0 when the youngest child is four years

old (or two years old in the control group). Hence, γt captures the general time trend.

The coefficients of interest are δt, which capture the average effect of the youngest child

period. This does not cause a time shock for the parents, since they still have a younger child athome. Parents do experience a small monetary shock when older children start going to school, sincethey pay for fewer hours of child care for these children during school hours. This monetary shock ishowever substantially smaller than the shock parents experience when the youngest child starts goingto school (see footnote 9). As a robustness check, we also restricted our control group to parentswhose second-youngest child is at least three years older to make sure no child in the household startsgoing to school within the observation period. This restriction severely limits the sample, but doesnot affect our results.

11This monetary shock is however substantially smaller than the shock parents experience when theyoungest child starts going to school, since own contributions for child care are generally highest forthe youngest child. The magnitude of the shock parents experience when their second-youngest childstarts going to school is shown in Appendix Table A1.

15

going to school for the treatment group (ATT) for each month relative to treatment.

Our reference category is 11 months before treatment. We include individual fixed

effects ηi to control for any individual-specific effects (e.g. the number of children in

the household, education level of the parents or ability) that might drive changes in

our outcome variable. We also include a full set of calendar year by month fixed effects

θt to control for any common time shocks (e.g. a recession). Finally, εit is the error

term and standard errors are clustered at the level of the parent to take into account

within-parent correlation in labor market behavior.

We also estimate a simpler version of equation 7 to get an average treatment effect

over the entire two-year post treatment period

yit = α + β × Ti + γt × St + δ × St × Ti + ηi + θt + εit, (8)

where St is a dummy equal to one if the youngest child is in school and and zero

otherwise and Ti is a dummy equal to one if the parent is in the treatment group and

zero otherwise. The average treatment effect on the treated is given by δ. All other

terms are defined as before. In the heterogeneity analyses we interact the treatment

dummy with indicators for different groups (e.g. marital status or number of children).

To ensure that we compare parents in the treatment and the control groups who

are as similar as possible, we apply coarsened exact matching (CEM) (Azoulay et al.,

2010; Iacus et al., 2012).12 Following Azoulay et al. (2010), we match on the pre-

treatment outcomes. In particular, we match on hours worked by a parent 11 and

6 months before the treatment and on the calendar year of treatment. The hours

a parent worked are divided in bins.13 In this way, we obtain weights for the control

groups that are proportional to the number of treated and controls in each combination

of (bins of) hours worked 11 and 6 months before treatment. All treated parents can

be matched to at least one parent in the control group. Only 0.0001% of the potential

control group is not matched.14

This matching procedure alleviates concerns regarding the common trend assump-

12In contrast to propensity score matching, CEM is nonparametric. CEM involves selecting co-variates on which we want to achieve balance, and then match each treated observation to controlobservations based on the values of the covariates. The approach is coarse in the sense that we do notmatch on each value for each covariate, but rather coarsen the distribution for some covariates andachieve balance on the coarsened distribution.

13The bins are defined in the following way: 0, 1-4, 5-9, 10-14, 15-19, 20-24, 25-29, 30-34, 35-39 and40 hours and above.

14Our results are comparable if we do not apply CEM.

16

tion, which is required in our difference-in-differences design. That is, conditional on

fixed effects and observables, trends in the treatment and control group should follow

a similar trend in the absence of treatment. The matching procedure we apply should

reduce possible differences in the pre-trend period. We also explicitly test for common

trends in outcomes during the pre-treatment period by estimating the 11 months before

treatment.

4.3 Data and descriptive statistics

We use administrative data on the universe of employees in the Netherlands.15 The

data consist of monthly employment records collected for the purpose of social security

administration and income taxes between 2006 and 2016. The data contain information

on actual hours worked and the wage for all jobs of a worker in that month.16 From

these administrative data, we construct a balanced panel of monthly data on parents’

average weekly working hours.17 We add data from the municipal registries (Gemeen-

telijke Basisadministratie) which contains demographic information on all households,

including for each household member the month and year of birth and gender.

We construct separate samples for mothers and their partners, who are the biologi-

cal fathers or the stepfathers to the children, independent of whether they are married

or not.18 We therefore do include single mothers in our sample, but not single fathers.

Subsequently we apply the following sample restrictions. First, we drop mothers

who have their first child before the age of fifteen or after the age of 40 (1%) as well as

fathers who have their first child before 15 or after 50 (1%), parents who work more than

100 hours per week during at least one month (0.02% of mothers and 0.09% of fathers),

parents who belong to two households at the same time (0% of mothers and 0.07% of

fathers) and finally mothers for whom the difference between their youngest and second-

youngest child is larger than 10 years (2.3%). We also drop observations for fathers

where couples divorce or separate during the observation period (7.5%) to ensure that

15The data are available to researchers who sign a confidentiality agreement with Statistics Nether-lands. While we cannot share the data, all programs to replicate our results are available on request.

16We use actual hours worked rather than contractual hours, since actual hours worked also takeparental leave into account. Parental leave policies differ substantially between sectors in the Nether-lands.

17Using the monthly data, we construct average hours worked per week by dividing the number ofhours worked in a month by 4.35 (the average number of weeks per month).

18Unfortunately same-sex couples cannot easily be identified in our data on household composition,because it is often unclear whether two people of the same gender are a couple or just living togetherin the same household. Hence we only include couples where the gender differs between the partners.

17

the partner remains the same. Finally, in the main analyses we exclude people who

receive income from self-employment at any point in the observation period, because

we do not observe their hours worked outside of hours spent on regular employment

(10.6% of the mothers and 19.5% of the fathers).

Table 2 shows the summary statistics for mothers (Columns 1 and 2) and fathers

(Columns 3 and 4) in the treatment and the control group weighted by matching

weights. Mothers in our sample are around 37 years old in the treatment group, and

as expected, somewhat younger in the control group. The treatment and the control

group have a similar share of highly educated mothers and ethnicity is also similar.

They have slightly more than 2 children on average. In the treatment group around

68% is married or in a civil union, while in the younger control group around 65%

is married. The age at which they gave birth to their first-born is very similar, as

is their hourly wage. Fathers are somewhat older than mothers, close to 40 years in

the treatment group and 38 in the control group. Ethnicity and number of children –

slightly more than 2 on average – are very similar. Education level is also very similar.

Close to 80% in the treatment group is married or is in a civil union, while this holds

for around 75% in the control group. The hourly wage is somewhat higher for the older

fathers in the treatment group.

Figure 3 reports descriptive statistics on the average number of hours mothers and

fathers worked and their employment status, both in the treatment and the control

group, weighted by matching weights. On average, mothers work almost 16 hours per

week (including those who do not work as zero). In addition, around 68% of the mothers

in our sample work in a given month pre-treatment. Finally, employed mothers work

around 23 hours on average per week. This suggests that there is substantial room

for increases along both the intensive and the extensive margin. It is clear that the

matching procedure succeeded in creating very similar groups in the pre-treatment

period. This should alleviate some concerns about the common trend assumption.

Around 90% of fathers are employed. Those who are employed, work close to 38

hours per week on average. The average number of hours fathers worked declined

somewhat during the three years they are in the sample. This decline is similar for

the treatment and control group and seems to be driven by a decline in employment

status. This could be due to the Great Recession in 2009 – 2010, and the second dip

the Netherlands experienced in 2011 – 2013.

18

Figure 3: Descriptive statistics on employment for mothers and for fathers

(a) Mothers’ weekly working hours(including zeros)

1415

1617

Wee

kly

hour

s w

orke

d

-12 -6 0 6 12 18 24Time relative to treatment in months

Treatment Control

(b) Fathers’ weekly working hours(including zeros)

3233

3435

Wee

kly

hour

s w

orke

d

-12 -6 0 6 12 18 24Time relative to treatment in months

Treatment Control

(c) Probability for mothers to be em-ployed

.65

.7.7

5Pr

obab

ility

to b

e em

ploy

ed

-12 -6 0 6 12 18 24Time relative to treatment in months

Treatment Control

(d) Probability for fathers to be em-ployed

.85

.9.9

5Pr

obab

ility

to b

e em

ploy

ed

-12 -6 0 6 12 18 24Time relative to treatment in months

Treatment Control

(e) Mothers’ weekly working hours ifemployed

2223

2425

Wee

kly

hour

s w

orke

d if

empl

oyed

-12 -6 0 6 12 18 24Time relative to treatment in months

Treatment Control

(f) Fathers’ weekly working hours ifemployed

3637

3839

Wee

kly

hour

s w

orke

d if

empl

oyed

-12 -6 0 6 12 18 24Time relative to treatment in months

Treatment Control

Notes: Own calculations based on register data from Statistics Netherlands. Treatment at t = 0. In the treatment

group, this is when the youngest child turns four years old and in the control group, this is when the youngest child

turns 2 years old.

Table 2: Descriptives of demographics for mothers and fathers in treatment and controlgroup weighted by matching weights.

Mothers Fathers

(1) (2) (3) (4)

Treatment group Control group Treatment group Control group

Age 36.80 34.65 39.97 37.90

(4.49) (4.58) (4.90) (5.06)

High educated 0.41 0.42 0.47 0.47

(0.49) (0.49) (0.50) (0.50)

Native 0.75 0.74 0.80 0.78

(0.43) (0.44) (0.40) (0.41)

Foreign born 0.17 0.18 0.14 0.15

(0.38) (0.39) (0.35) (0.36)

Native born from foreign-born parent 0.07 0.08 0.06 0.07

(0.26) (0.27) (0.24) (0.25)

Number of children in the household 2.15 2.17 2.18 2.16

(0.88) (0.90) (0.84) (0.86)

Not married 0.20 0.24 0.20 0.25

(0.40) (0.43) (0.40) (0.43)

Married 0.68 0.66 0.80 0.75

(0.47) (0.47) (0.40) (0.43)

Single parent 0.12 0.10

(0.33) (0.30)

Age at first born 28.98 28.80 32.23 32.21

(4.73) (4.76) (5.11) (5.20)

Hourly wage pre-treatment 17.01 16.62 21.65 20.66

(10.97) (10.39) (17.29) (16.59)

Hourly wage year 1 post-treatment 17.13 16.80 21.99 21.08

(10.77) (9.79) (21.41) (16.45)

Hourly wage year 2 post-treatment 17.30 17.04 22.36 21.48

(11.73) (10.44) (54.72) (15.80)

Observations 21,473,784 21,809,088 15,364,404 16,056,612

No. of individuals 596,494 605,808 426,789 446,017

Notes: The table reports means and standard deviations in parentheses. We observe all variables for the full sample,

except for education, which is only observed for about two-thirds of the sample and is biased towards higher-educated.

5 Results

In Section 3, we calculated that the treatment mothers experience when their youngest

child starts going to school is a time shock of on average 13.3 hours per week and a

monetary shock to the household of on average 61 euros per month. Similarly, fathers

and stepfathers on average experience a time shock of 3.7 hours and a monetary shock

to the household of 70 euros. Furthermore, Section 4 showed based on descriptive

statistics that employment among mothers seems to increase after treatment has taken

20

place, while the treatment hardly seems to affect men. In this section, we show the

results of the difference-in-differences analyses, which confirm the descriptive results.

5.1 Main results

Figure 4 reports estimates of the effect of the youngest child going to school (at t = 0)

on maternal and paternal employment. We present estimates for each month from the

year before until two years after the child goes to school. The estimates are relative to

the situation eleven months before the child goes to school.

Figure 4a reports the estimated differences between individuals in the treatment

and the control groups on total working hours, including those who do not work as

zero. The precisely estimated average effect is a small increase in weekly working hours

after the youngest child goes to school of close to 0.5 hours per week after two years.

This is a 3% increase relative to the pre-treatment mean of around 15 hours per week of

the mothers in the treatment group. Over the entire post-treatment period the average

effect is 0.3 hours, or 1.9% relative to the mean.

About 32% of the mothers in our sample does not work before their child goes to

school. It is possible that the time that they no longer need to spend on childcare,

induces some of them to enter the labor force. Figure 4c reports effects on the probabil-

ity to work. We find an increase of about 1.5 percentage points relative to the control

group after two years in the chance that a mother works. This is a 2.2% increase

relative to the pre-school mean of 69% of mothers who work. The average treatment

effect is 0.9 percentage points, or 1.3% relative to the mean.

Finally, Figure 4e reports effects on weekly working hours for those in employment

throughout the entire observation period. We find that those in employment work

about 0.4 additional hours per week after their child goes to school. The figures provide

evidence that our matching procedure succeeded in making our treatment and control

groups comparable. We find no evidence of changes in labor supply in the pre-treatment

period. This implies that there does not seem to be an anticipation effect.

We can now calculate the contribution of the changes at the intensive and extensive

margin to the total effect. The contribution of the effect at the intensive margin is equal

to the share who work multiplied by the effect size 0.68× 0.3 = 0.2. The total effect is

0.3, which means that the effect measured in hours worked per week at the extensive

margin must be 0.1. This means that changes along the intensive margin contribute

about two-thirds to the total effect, while changes along the extensive margin contribute

21

about one-thirds. Furthermore, given that we find an increase along the extensive of

0.9 percentage point, this means that mothers who start working when their youngest

child goes to school on average work around 11 hours (0.1/0.009). This is about half the

number of hours worked by those who already work before their child goes to school.

Next, we consider how the youngest child in a household going to school affects the

partners of these mothers, if they have one during the observation period. Figure 4

shows that on average fathers and stepfathers respond much less than mothers do to

their child going to school. We find that weekly working hours increase by about 0.15

hours on average, or a 0.4% increase relative to the mean of 34 hours before treatment.

The probability to work also increases slightly by about 0.5 percentage points, or a

0.6% increase relative to the mean of 91% before treatment. Finally, the effect on

hours worked conditional on employment is very close to zero for fathers. For fathers

the effect along the intensive margin (0.91× 0.05 = 0.05) contributes about one-thirds

to the total effect, while the effect along the extensive margin contributes about two-

thirds. This means that fathers who start working after their youngest child goes to

school, work about 20 hours on average (0.1/0.005 = 20). However, while statistically

significant, the effects are very small. Also note that for fathers we observe a small

pre-trend, suggesting that other factors could also play a role here.

Overall, these results show that the treatment leads to a small increase in time spent

at work for mothers and an even smaller increase for fathers. The increase for mothers

is both along the intensive and the extensive margin. At the same time, although

the effects are statistically significant, they are economically quite small. They are

also small compared to the magnitude of the change in time mothers experience when

their youngest child starts going to school. Fathers generally experience a substantially

smaller shock, particularly in terms of time, when their youngest child starts going to

school than mothers do. This may contribute to the finding that mothers respond more

than fathers when the youngest child in a household starts going to school. We look

into this in the next section.

5.2 Heterogeneity

On average the treatment involves an increase of 13.3 hours of additional time and

61 euros of income per month for mothers and an increase of 3.7 hours and 70 euros

for fathers. There is however important heterogeneity in the treatment as was shown

in Table 1. Parents who work more hours gain less additional time than parents who

22

Figure 4: Main estimates for mothers and fathers

(a) Effect on weekly working hours formothers (including zeros)

-.50

.51

1.5

22.

5C

hang

e in

wee

kly

hour

s w

orke

d

-12 -6 0 6 12 18 24Time relative to treatment in months

Average treatment effect: 0.31*** (0.01)

(b) Effect on weekly working hours for fa-thers (including zeros)

-.50

.51

1.5

22.

5W

eekl

y ho

urs

wor

ked

-12 -6 0 6 12 18 24Time relative to treatment in months

Average treatment effect: 0.14*** (0.02)

(c) Effect on probability to work for moth-ers

-20

24

68

%-p

oint

cha

nge

in p

roba

bilit

y to

be

empl

oyed

-12 -6 0 6 12 18 24Time relative to treatment in months

Average treatment effect: 0.90*** (0.04)

(d) Effect on probability to work for fa-thers

-20

24

68

%-p

oint

cha

nge

in p

roba

bilit

y to

be

empl

oyed

-12 -6 0 6 12 18 24Time relative to treatment in months

Average treatment effect: 0.25*** (0.04)

(e) Effect on weekly working hours if em-ployed for mothers

-.50

.51

1.5

22.

5C

hang

e in

wee

kly

hour

s w

orke

d if

empl

oyed

-12 -6 0 6 12 18 24Time relative to treatment in months

Average treatment effect: 0.31*** (0.01)

(f) Effect on weekly working hours if em-ployed for fathers

-.50

.51

1.5

22.

5W

eekl

y ho

urs

wor

ked

if em

ploy

ed

-12 -6 0 6 12 18 24Time relative to treatment in months

Average treatment effect: 0.05*** (0.01)

Notes: Own calculations based on register data from Statistics Netherlands. Treatment at t = 0. For the treatment

group this is when the youngest child turns four years old. For the control group this is when the youngest child turns

2 years old. The lighter band around the estimate represents the 99% confidence interval. Reported average treatment

effect is based on Equation 8. Cluster-robust standard errors clustered by individual in parentheses, ∗∗∗ p<0.001, ∗∗

p<0.01, ∗ p<0.05. The coefficients for probability to work are multiplied by 100 so that they can be interpreted as

percentage point changes.

work very few hours per week. Additionally, households in which the parents earn

more, gain more money when their youngest child goes to school than household with

a lower household income because childcare subsidies are income dependent.

Theoretically, we would expect larger positive responses in labor supply for those

with the largest time gains and the smallest financial gains. However, our heterogeneity

analyses come to different conclusions. Differences in initial preferences for labor supply

are probably at the heart of these results.

Table 3 shows the differences in the estimated effects on the hours worked (including

those who do not work) for mothers (Column 1) and fathers (Column 3) who work

different numbers of hours (Panel A) and who earn different wages per hour (Panel

B) in the year before treatment. Column 2 (Column 4) reports the mean number of

hours mothers (fathers) in the treatment group worked in the twelve months before

their youngest child started going to school. Panel A shows that –surprisingly– the

increase in hours worked is largest for mothers who already worked more than 34 hours

before treatment. The increase in hours worked is smallest among those who did not

work before the treatment. So although the average number of hours mothers worked

increases significantly amongst all groups, mothers who obtain more additional time

(because they work fewer hours) when their youngest child starts going to school in fact

increase their working hours less than mothers who only obtain very little additional

time (because they already worked (nearly) fulltime before the treatment).

On average, fathers experience a substantially smaller shock in terms of time when

their youngest child starts going to school. Column 3 of Table 3 illustrates that fathers

who do experience a larger shock in terms of time (because they work few hours prior

to the treatment), also do not significantly increase their working hours.These findings

suggest that there are reasons other than caring for the youngest child to explain why

these fathers do not work, such as disability or study. These results point out that

the differences between the estimated average treatment effects between mothers and

fathers are not simply entirely due to differences in the magnitude of the shock they

experience when their youngest child goes to school. Even within groups in which

the number of hours parents work is relatively similar, mothers increase their working

hours more or at least as much as fathers do.

Panel B reports the heterogeneity by the hourly wage parents earn. In Section 4,

we formulated the hypothesis that for parents who earn higher wages, the effects are

larger. Clearly wages are only defined for parents who are working. For mothers, we

24

Table 3: Heterogeneity by hours worked and wage quartile.

Mothers Fathers

(1) (2) (3) (4)

Estimated effect Avg hours worked Estimated effect Avg hours worked

Panel A. Heterogeneity by hours worked

Not working 0.22∗∗∗ 0.0 −0.01 0.0

(0.02) (0.09)

1-20 hours 0.35∗∗∗ 12.9 0.12 10.8

(0.02) (0.14)

20-33 hours 0.31∗∗∗ 25.5 0.31∗∗∗ 29.9

(0.01) (0.05)

34+ hours 0.50∗∗∗ 37.9 0.13∗∗∗ 38.8

(0.06) (0.01)

Observations 43,282,872 31,421,016

Panel B. Heterogeneity wage quartile

Quartile 1 0.22∗∗∗ 19.1 0.06 36.1

(0.03) (0.03)

Quartile 2 0.36∗∗∗ 21.3 0.08∗∗∗ 37.1

(0.02) (0.02)

Quartile 3 0.44∗∗∗ 22.4 0.10∗∗∗ 36.8

(0.02) (0.02)

Quartile 4 0.45∗∗∗ 25.5 0.24∗∗∗ 37.0

(0.02) (0.03)

Observations 29,785,932 28,631,088

Panel C. Full sample

Average treatment effect 0.31∗∗∗ 15.6 0.14∗∗∗ 33.9

(0.01) (0.02)

Observations 43,282,872 31,421,016

Notes: The table reports the total estimated effect for each group from a regression with interactions between

the treatment indicator and indicators for each groups. Cluster-robust standard errors clustered by individual

in parentheses, ∗∗∗ p<0.001, ∗∗ p<0.01, ∗ p<0.05. All regressions include individual fixed effects and calendar

year-month fixed effects.

see that women who earn higher wages on average work more hours, while this does not

seem to be the case for fathers. We find that all groups increase their hours worked,

and that for both mothers and fathers, the increase is indeed largest among those who

earn higher hourly wages.

Appendix Table A2 additionally reports how the effects differ by demographic char-

acteristics. Contrary to what is commonly found in the literature, we see in Panel

A that single mothers actually decrease the number of hours they work when their

youngest child starts going to school. This might be explained by differences in the

25

examined time periods: most papers on the effect of compulsary schooling on maternal

labor supply use data from the 1980s and the 1990s and in those days the composition

and preferences of the female labor force were distinctly different. At the same time,

the problem might also be that our control group is not suitable for single mothers:

perhaps single mothers with very young children are inherently different from single

mothers with slightly older children.

Panel C shows that mothers with one child show a small decline in hours worked

and the probability to work, while mothers with more children increase both their

hours and participation in the labor market. One explanation could be that mothers

with more children wait until their youngest is off to school, and only then enter the

labor market, while mothers with only one child were already working more. Indeed,

the data show that mothers with one child work 17.7 hours on average, while mothers

with more than two children work 16.9 hours on average and mothers with three or

more children work 11.6 hours on average.

5.3 Robustness analyses

In the main analyses, we did not include self-employed workers in our sample, because

we do not observe the number of hours they worked. Because this restriction reduces

the sample by a sizable 10.6% for mothers and 19.5% for fathers, it is important to

discuss whether this restriction affects our results.

To check this, we estimate the hours self-employed mothers (fathers) worked in

a given year by dividing the profits they reported by the average hourly wage of all

mothers (or fathers) in our sample in a given calendar year.19 Clearly using the average

hourly wage to calculate the number of hours self-employed parents work yields only

a rough estimation of the actual number of hours a self-employed parent works. Still,

the estimates for the probability to work are reliable. Other than excluding the self-

employed parents, we apply the same sample selections as in the main analysis.20

19We therefore assume that these self-employed parents are working in each month throughout thecalendar year if they have income from self-employment in that year. If self-employed workers arealso employed at some point in time, we simply sum the hours worked as self-employed and in theiremployment.

20We again drop people who work more than 100 hours per week in either a job, or as self-employedor as a combination of both. For mothers this is 0.7% of the sample and for fathers it is 1.5%. Wethen run the same matching procedure as described in the main text. We end up with 1,341,925matched mothers and 1,127,896 matched fathers. This is 11.6% (29.2%) more than the main sample.66% (91%) of the mothers (fathers) in this sample work in a given month before treatment, and theywork 16.3 (33.7) hours on average.

26

Descriptives are reported in Table A3 in the Appendix.

Table 4 illustrates that including self-employed in the sample does not change the

main conclusions. Panel C reproduces the original estimates from the main analyses

and Panel A shows how the effects change when self-employed people are included in

the sample. For mothers, we find effects very similar to the main estimates. Including

self-employed, mothers work about 1.8% more hours on average in the two years after

their youngest child starts going to school, compared to 2% for the main estimates. The

estimates for hours worked for those who are employed and for the extensive margin are

also remarkably similar. For fathers, we find an average increase in total hours worked

of 0.3% relative to the mean, compared to an increase of 0.4% for the main estimates.

We find no statistically significant effect for those in work, which means that the full

increase is driven by the extensive margin. The results on the extensive margin are

very similar to the main estimates. This suggests that the substantial sample selection

of self-employed does not bias our results.

5.3.1 An alternative control group

To further check the robustness of our results, we also consider an alternative control

group. In our main analyses, our control group consists of parents whose youngest child

is between one and four years old. Their youngest child does not go to school in the

observation period. An alternative approach is to use parents whose second-youngest

child is between three and six years old as a control group.21 In line with the parents in

the control group in our main analyses, parents in this alternative control group do not

experience a time shock during the observation period, because they need to take care

of their youngest child who does not go to school yet. Parents in this alternative control

group, however, do experience a small monetary shock, because they no longer need to

pay for child care in the hours during which their second-youngest child now starts going

to school. In the Netherlands, government subsidies are substantially higher for the

child in the household that uses most hours of child care (generally the youngest child)

than for other children. As a result, the monetary shock households experience when

their second-youngest child starts going to school are substantially lower. Appendix

Table A1 shows the distribution of the magnitude of the shock fathers and mothers in

this alternative control group experience.

21To ensures that parents in this group do not experience a time shock because their youngest childstarts going to school while they are in this control group, the youngest child should be two or threeyears younger and the child remains the second-youngest child in the observation period.

27

We apply the same selections as in the baseline estimates. We end up with a sample

of 596,494 (426,789) treated mothers (fathers) and 152,947 (122,012) control mothers

(fathers). Descriptives are reported in Table A4 in the Appendix.

Panel B in Table 4 reports the estimation results. For mothers we find somewhat

smaller effects on average, but larger effects along the intensive margin. We find no

effect along the extensive margin. For fathers we actually find small negative effects on

hours worked overall. These appear to be driven by the extensive margin. However,

while statistically significant, the results are, just as the baseline estimates, economi-

cally insignificant: we find a decline in hours worked overall of 0.5% for fathers. Hours

worked while employed increases by 0.05 hours, or 0.1%, just as in the baseline esti-

mates. We also find a small decline in probability to work of about 0.6%.

28

Table 4: Treatment effect estimates on hours per week, hours worked if employed and the probability to work whenincluding self-employed and relative to a control group consisting of parents of children with a second-youngest child aged3–6.

Mothers Fathers

(1) (2) (3) (4) (5) (6)

Hours workedHours worked if

working

Probability to

workHours worked

Hours worked if

working

Probability to

work

Panel A. Including self-employed

Average treatment effect 0.29∗∗∗ 0.28∗∗∗ 0.87∗∗∗ 0.10∗∗∗ 0.02∗ 0.24∗∗∗

(0.01) (0.01) (0.04) (0.02) (0.01) (0.04)

Observations 48,309,300 26,732,808 48,309,300 40,604,256 31,022,640 40,604,256

Mean dep. variable 16.2 23.7 0.70 33.7 37.5 0.91

Panel B. Relative to control group with second-youngest child 3-6

Average treatment effect 0.23∗∗∗ 0.45∗∗∗ 0.09 −0.15∗∗∗ 0.08∗∗∗ −0.54∗∗∗

(0.02) (0.01) (0.07) (0.02) (0.01) (0.06)

Observations 26,979,876 15,414,624 26,979,876 19,756,836 15,759,468 19,756,836

Mean dep. variable 15.6 23.2 0.69 33.9 37.6 0.91

Panel C. Main results

Average treatment effect 0.31∗∗∗ 0.31∗∗∗ 0.90∗∗∗ 0.14∗∗∗ 0.05∗∗∗ 0.25∗∗∗

(0.01) (0.01) (0.04) (0.02) (0.01) (0.04)

Observations 43,282,872 23,990,148 43,282,872 31,421,016 24,695,748 31,421,016

Mean dep. variable 15.6 23.2 0.69 33.9 37.6 0.91

Notes: Cluster-robust standard errors clustered by individual in parentheses, ∗∗∗ p<0.001, ∗∗ p<0.01, ∗ p<0.05. All regressions include individual fixed effects and

calendar year-month fixed effects. The treatment group consists of mothers with a youngest child 3–6 years old. The control group in Panels A and C consists of

mothers with a youngest child between 1 and 4 years old. In Panel B the control group consists of parents with a second-youngest child 3–6 years old. The coefficients

for probability to work are multiplied by 100 so that they can be interpreted as percentage point changes.

6 Conclusion and discussion

This paper shows that when the youngest child in a family starts going to school, moth-

ers in the Netherlands only increase their labor supply marginally. When the youngest

child starts attending school, mothers experience an increase in free time of more than

thirteen hours a week, which is expected to increase their labor supply. Theoretically,

this effect may be mitigated by the income effect that households experience simul-

taneously: when the youngest child starts attending school, parents save on average

sixty to seventy euros on the costs of formal child care. Nonetheless, even though such

income effects are expected to decrease parental labor supply, the effects are likely to

be small in this case, because the monetary shock is relatively small.

Compared to mothers whose youngest child turned two and who therefore do not

experience any shock in the observation period, mothers increase their labor supply

by on average approximately 0.5 hours a week after two years. This is around a 3%

increase relative to the mean before their child goes to school. For their partners we

find an even smaller increase of about 0.15 hours after two years, or 0.4% relative to the

mean before their child goes to school. These findings are generally robust to alternative

specifications. However, employing an alternative control group of parents with second-

youngest children leads to small negative estimates for partners and somewhat smaller

positive effects for mothers.

Our heterogeneity analyses shows surprising results: the response is larger among

mothers and fathers who are already working longer hours and who already earn higher

wages. Theoretically, one would expect smaller increases in working hours in these

groups as the relative gain in time and opportunity costs of child care are smaller.

However, differences in initial preferences for labor supply over leisure are probably

driving these results.

Utility or consumption smoothing can perhaps be an explanation for the relatively

limited effects that we find in this paper. Clearly, the moment a child turns four

is known in advance. Assuming that parents wish to keep their marginal utility of

consumption and leisure as similar as possible over time, only small changes to the

number of hours worked can be expected. An alternative explanation is that parents

in fact consider the gross monetary gain - the difference in the cost of child care

without considering the corresponding decline in child care subsidies - they obtain

when their youngest child starts going to school, rather than the net monetary gain

that we calculate in this paper. In the Netherlands parents receive child care subsidies

30

from the tax authorities and often on a different date than when they pay for child

care. The gross monetary gain is substantially larger than the net monetary gain,

especially for parents with a relatively low household income, and this leads to a larger

perceived increase in income than is actually the case. In addition, the literature

shows that successive cohorts of parents respond less and less strongly to incentives

aimed at increasing their labor participation. Finally, it could be that school times are

inconvenient, making it difficult to increase hours worked during school hours.

31

References

Abeler, J. and F. Marklein (2016). Fungibility, labels, and consumption. Journal of

the European Economic Association 15 (1), 99–127.

Azoulay, P., J. S. Graff Zivin, and J. Wang (2010). Superstar extinction. The Quarterly

Journal of Economics 125 (2), 549–589.

Blau, F. D. and L. M. Kahn (2007). Changes in the labor supply behavior of married

women: 1980–2000. Journal of Labor Economics 25 (3), 393–438.

Blundell, R., A. Bozio, and G. Laroque (2013). Extensive and intensive margins of

labour supply: Work and working hours in the US, the UK and France. Fiscal

Studies 34 (1), 1–29.

Cascio, E. U. (2009). Maternal labor supply and the introduction of kindergartens into

American public schools. Journal of Human Resources 44 (1), 140–170.

Devereux, P. J. and R. A. Hart (2010). Forced to be rich? Returns to compulsory

schooling in Britain. The Economic Journal 120 (549), 1345–1364.

Eurostat (2012). Being young in europe today: Education.

Finseraas, H., I. Hardoy, and P. Schøne (2016). School enrolment and mothers’ labor

supply: evidence from a regression discontinuity approach. Review of Economics of

the Household , 1–18.

Fitzpatrick, M. D. (2010). Preschoolers enrolled and mothers at work? the effects of

universal prekindergarten. Journal of Labor Economics 28 (1), 51–85.

Fitzpatrick, M. D. (2012). Revising our thinking about the relationship between ma-

ternal labor supply and preschool. Journal of Human Resources 47 (3), 583–612.

Gelbach, J. B. (2002). Public schooling for young children and maternal labor supply.

The American Economic Review 92 (1), 307–322.

Goux, D. and E. Maurin (2010). Public school availability for two-year olds and moth-

ers’ labour supply. Labour Economics 17 (6), 951 – 962.

32

Grenet, J. (2013). Is extending compulsory schooling alone enough to raise earnings?

Evidence from French and British compulsory schooling laws. The Scandinavian

Journal of Economics 115 (1), 176–210.

Iacus, S. M., G. King, and G. Porro (2012). Causal inference without balance checking:

Coarsened exact matching. Political Analysis 20 (1), 1–24.

Kinderopvang, B. (2016). Factsheet kinderopvang.

Machin, S., O. Marie, and S. Vujic (2011). The crime reducing effect of education. The

Economic Journal 121 (552), 463–484.

Rijksoverheid (2017). Overzicht aantal uren onderwijstijd.

Shure, N. (2019). School hours and maternal labor supply. Kyklos 72 (1), 118–151.

Thaler, R. (1985). Mental accounting and consumer choice. Marketing Science 4 (3),

199–214.

33

A Appendix

A.1 Description of maternal budget constraint

This appendix explains in more detail how the time mothers spend on work, taking care

of their children and leisure changes when the youngest child starts going to school,

i.e. when ts > 0. Figure 2 shows the budget constraint mothers face when at least one

child in a household does not go to school and the budget constraint mothers face when

the youngest child in a household starts going to school. In this figure the time fathers

spend on labor lf and child care tf is held constant. Mothers decide how to allocate

their time between leisure zm, taking care of their children tm and working from which

they earn an income that enables them to consume X. The blue solid planes represent

the situation when at least one child does not go to school yet and the green dashed

planes represent the situation after the youngest child has started going to school when

parents no longer have to arrange child care during school hours.

Figure 2a demonstrates the tradeoff between the time mothers take care of their

children tm and the time they spend on leisure zm, Figure 2b depicts the tradeoff be-

tween the mother’s leisure zm and the household’s consumption X and Figure 2c shows

the relation between the time mothers spend on domestic child care and household con-

sumption.

In this section we explain the budget constraint by zooming in on four situations

that correspond to four points on this budget constraint. A first situation of interest

is when mothers work fulltime and therefore do not enjoy any free time and do not

provide domestic child care (1: lm = 1, zm = 0, tm = 0). A second situation is when

mothers spend all their time on leisure (lm = 0, zm = 1, tm = 0). At the third and

fourth situation we discuss, the mother spends as much time as she can on domestic

child care. When a child goes to school or when the father (whose time use is held

constant in Figure Figure 2) also provides child care, mothers cannot spend all their

time on domestic child care. The third situation therefore entails that women spend as

much time as possible on providing domestic child care and spend the remainder of their

time working (3A: lm = ts + tf , zm = 0, tm = 1− ts− tf ). Finally in the fourth situation

the mother again maximizes the time she spends on providing domestic child care and

spends the remainder of her time on leisure (3B: lm = 0, zm = ts + tf , tm = 1− ts− tf ).

First, when mothers work fulltime and do not enjoy any free time and do not

provide domestic child care (1: lm = 1, zm = 0, tm = 0), the household pays for child

34

care while the parents work and consumes the earnings of the father and the mother.

Household consumption therefore equals X = w∗f lf + w∗

m − q∗(1 − ts − tf ), where the

asterisks denote real wages and real prices of child care, w∗i = wi/p and q∗ = q/p. In

Figure 2a this point is at the origin: mothers do not provide child care nor do they

enjoy leisure. In Figure 2b and in Figure 2c this situation is at the top-left part of the

budget constraint. As a result, there is a linear upward shift of the budget constraint

from the blue plane to the green dashed plane at this point. When the youngest child

in a household starts going to school, household consumption increases by q∗ts and

depending on the shape of the utility function this causes mothers to increase their

leisure and thereby decrease their labor supply.

A second situation of interest is when mothers spend all their time on leisure. In

this situation, women do not work but children attend external child care outside of

possible school hours and when their father does not take care of them. Household

consumption at this point is therefore X = w∗f lf − q∗(1 − ts − tf ). In line with the

previous situation, there is an upward linear shift of the budget constraint when the

youngest child starts going to school and parents only experience an income effect.

Third, when mothers spend as much time as they can on providing domestic child

care, they additionally experience a time effect. In situation 3A, mothers maximize the

time they spend on child care and work the remainder of the time, lm = ts + tf , zm =

0, tm = 1 − ts − tf . In this situation, children do not attend external child care.

Household consumption then equals X = w∗f lf +w∗

m(ts + tf ). When the youngest child

starts attending school, the time these mothers spend on domestic child care decreases

from 1− tf to 1− ts − tf and her labor supply increases correspondingly.

Finally, in situation 3B mothers spend as much time as possible on providing domes-

tic child care and spend the remainder of the time on leisure, lm = 0, zm = ts + tf , tm =

1− ts − tf . Household consumption in this situation is equal to X = w∗f lf as children

do not attend external child care and mothers do not work. When the youngest child

starts going to school, these mothers spend less time on domestic child care and enjoy

more leisure.

In sum, when the youngest child starts going to school, the budget constraint

changes in two ways: on the one hand there is an upward linear shift because mothers

whose children attend external child care experience an income effect as their child

care expenses decrease. On the other hand, the maximum amount of time mothers can

spend on domestic child care decreases. The exact optimal situation in the situation

35

Table A1: Magnitude of the shock in terms of money and in terms of time

Panel A: Shock in terms of time and money that mothers experience

when their second-youngest child goes to school

(1) (2) (3)

% of Hours saved Money saved

mothers per week per month

(per household)

Does not work 19.2 0 3

Works less than 20 hours a week 27.3 0 11

Works 20-34 hours a week 48.6 0 28

Works more than 34 hours a week 4.9 0 48

Total 100.0 0 19

Panel B: Shock in terms of time and money that fathers experience

when their second-youngest child goes to school

% of Hours saved Money saved

fathers per week per month

(per household)

Does not work 4.8 0 9

Works less than 20 hours a week 3.4 0 14

Works 20-34 hours a week 18.1 0 23

Works more than 34 hours a week 73.6 0 21

Total 100.0 0 21

Notes: See main text for calculation of the monetary and time shocks. The percentages listed

in Column (1) refer to the share of mothers (fathers) in our sample that works a specific number

of hours. The amounts of money households save when their second-youngest child starts going

to school (listed in Column (3)) differs for the mothers (Panel A) and fathers (Panel B) in our

sample, because households with single mothers are included in the sample for mothers, but not

in the sample of fathers.

before and after the youngest child starts going to school depends on the marginal

utility she obtains from those activities, and thereby on the shape of the utility function.

A.2 Additional results

36

Table A2: Heterogeneity by marital status, ethnicity and number of children.

Mothers Fathers

(1) (2) (3) (4)

Estimated effect Avg hours worked Estimated effect Avg hours worked

Panel A. Heterogeneity by marital status

Dual household 0.37∗∗∗ 15.9

(0.01) (0.01)

Single household −0.29∗∗∗ 13.2

(0.05) (0.04)

Panel B. Heterogeneity by ethnicity

Native 0.40∗∗∗ 16.8 0.07∗∗∗ 35.1

(0.01) (0.02)

Foreign 0.05∗ 12.1 0.30∗∗∗ 29.6

(0.02) (0.04)

Panel C. Heterogeneity by number of children

1 child −0.17∗∗∗ 17.7 0.18∗∗∗ 33.2

(0.03) (0.04)

2 children 0.45∗∗∗ 16.9 0.14∗∗∗ 34.7

(0.01) (0.02)

3 or more children 0.41∗∗∗ 11.6 0.10∗∗∗ 32.7

(0.02) (0.03)

Panel D. Full sample

Average treatment effect 0.31∗∗∗ 15.6 0.14∗∗∗ 33.9

(0.01) (0.02)

Observations 43,282,872 31,421,016

Notes: The table reports the total estimated effect for each group from a regression with interactions between

the treatment indicator and indicators for each groups. Cluster-robust standard errors clustered by individual

in parentheses, ∗∗∗ p<0.001, ∗∗ p<0.01, ∗ p<0.05. All regressions include individual fixed effects and calendar

year-month fixed effects.

Table A3: Descriptives of demographics for mothers and fathers in treatment andcontrol group weighted by matching weights for the sample including self-employed.

Mothers Fathers

(1) (2) (3) (4)

Treatment group Control group Treatment group Control group

Hours worked per week pre-treatment 16.26 16.20 33.58 33.46

(13.58) (13.42) (14.02) (13.93)

Fraction working pre-treatment 0.70 0.70 0.91 0.91

(0.46) (0.46) (0.29) (0.29)

Hourly wage pre-treatment 17.15 16.77 22.43 21.33

(11.61) (11.54) (25.53) (19.80)

Hourly wage year 1 post-treatment 17.25 16.95 22.69 21.66

(10.99) (11.01) (25.87) (18.95)

Hourly wage year 2 post-treatment 17.42 17.17 22.98 22.02

(12.33) (11.06) (53.92) (18.87)

Self-employed at some point 0.11 0.10 0.23 0.22

(0.31) (0.30) (0.42) (0.41)

Age 36.91 34.77 40.06 38.03

(4.47) (4.56) (4.89) (5.06)

Fraction high educated 0.43 0.44 0.47 0.47

(0.49) (0.50) (0.50) (0.50)

Native 0.76 0.74 0.80 0.79

(0.43) (0.44) (0.40) (0.41)

Foreign born 0.17 0.18 0.13 0.14

(0.37) (0.38) (0.34) (0.35)

Native born from foreign-born parent 0.07 0.08 0.06 0.07

(0.26) (0.27) (0.24) (0.25)

Number of children 2.16 2.17 2.21 2.20

(0.88) (0.90) (0.86) (0.88)

Not married 0.20 0.25 0.21 0.26

(0.40) (0.43) (0.41) (0.44)

Married 0.68 0.66 0.79 0.74

(0.46) (0.47) (0.41) (0.44)

Single parent 0.12 0.09

(0.32) (0.29)

Age at first born 29.08 28.93 32.24 32.25

(4.71) (4.76) (5.15) (5.25)

Observations 24,117,624 24,191,676 20,004,552 20599704

No. of individuals 669,934 671,991 555,682 572,214

Notes: The table reports means and standard deviations in parentheses. We observe all variables for the full

sample, except for education, which is only observed for about two-thirds of the sample and is biased towards

higher-educated.

Table A4: Descriptives of demographics for mothers and fathers in treatment andcontrol group weighted by matching weights for the sample using a control group ofparents whose second-youngest child is between three and six years old.

Mothers Fathers

(1) (2) (3) (4)

Treatment group Control group Treatment group Control group

Hours worked per week pre-treatment 15.67 15.71 33.89 33.90

(12.65) (12.57) (12.29) (12.11)

Fraction working pre-treatment 0.69 0.69 0.91 0.91

(0.46) (0.46) (0.29) (0.29)

Hourly wage pre-treatment 17.01 17.52 21.71 21.47

(10.97) (11.81) (17.14) (16.52)

Hourly wage year 1 post-treatment 17.13 17.73 21.99 21.87

(10.77) (10.32) (21.41) (18.71)

Hourly wage year 2 post-treatment 17.30 18.00 22.35 22.32

(11.73) (12.10) (56.77) (15.35)

Age 36.80 34.17 39.97 37.02

(4.49) (4.13) (4.90) (4.58)

High educated 0.41 0.52 0.47 0.55

(0.49) (0.50) (0.50) (0.50)

Native 0.75 0.80 0.80 0.84

(0.43) (0.40) (0.40) (0.37)

Foreign born 0.17 0.13 0.14 0.10

(0.38) (0.34) (0.35) (0.30)

Native born from foreign-born parent 0.07 0.07 0.06 0.06

(0.26) (0.25) (0.24) (0.24)

Number of children 2.15 2.38 2.18 2.31

(0.88) (0.75) (0.84) (0.66)

Not married 0.20 0.24 0.20 0.25

(0.40) (0.43) (0.40) (0.43)

Married 0.68 0.71 0.80 0.75

(0.47) (0.45) (0.40) (0.43)

Single parent 0.12 0.05 0.00 0.00

(0.33) (0.21) (0.00) (0.00)

Age at first born 28.98 28.97 32.23 32.06

(4.73) (4.22) (5.11) (4.53)

Observations 21,473,784 5,506,092 15,364,404 43,924,32

No. of individuals 596,494 152,947 426,789 122,012

Notes: The table reports means and standard deviations in parentheses. We observe all variables for the full

sample, except for education, which is only observed for about two-thirds of the sample and is biased towards

higher-educated.